The Research on autopilot system based on lightweight YOLO-V3 target detection algorithm

Abstract

Abstract Computer vision is one of the most important branches in the field of deep learning, which includes image classification, target detection and so on. In this paper, the pruning algorithm is combined with the method of reducing the network depth, and the trained network model is applied to the field of autopilot, which solves the problems of high power consumption, single function and poor real-time performance of the vehicle target detection system. In this paper, through a large number of theoretical research and experiments, the appropriate activation function and regularization method are selected. Through the visualization method of neural network convolution layer, the working principle of the network is expounded in detail. Through theoretical analysis, the appropriate anchor box, is selected and the non-maximum suppression algorithm is introduced. Tensorboard data visualization system is used to monitor the distribution and law of network parameters. The accuracy of detection according to the experimental target is 98.7%.